v0.6.4

Highlights

• Fixed live Hermes profile evaluation so generated agents run from the generated profile directory instead of inheriting the template author's repo context.
• Added isolated live-profile installation, configurable provider/model/timeout/max-turn controls, and cleanup for noisy Hermes CLI warning output.
• Stopped generated customer profiles from bundling template authoring skills, keeping generated agents focused on their domain workflow.
• Reworked the local web demo into a finite-state interface with discrete prompt, progress, and fullscreen results panels plus smooth transitions and back/start-over navigation.

Verification

• PYTHONDONTWRITEBYTECODE=1 python3 -m py_compile scripts/*.py web-demo/server.py
• python3 scripts/validate_profile.py .
• make generate-smoke
• make sentence-smoke
• make smoke
• make release-check BASE=origin/main
• git diff --check
• Temporary HERMES_HOME install smoke with hermes profile install . --name release-smoke --yes --force
• Temporary HERMES_HOME hermes profile info release-smoke
• Authenticated temporary HERMES_HOME chat smoke with hermes -p release-smoke chat --provider openai-codex -m gpt-5.5 -Q --max-turns 3 -q ...
• Local web-demo HTTP smoke confirming the finite-state panel flow and generated file output.